AI Finds 33 Overlooked 2D Material Candidates in Decades-Old Crystal Data

Compounds synthesized in the 1970s, with their crystal structures already on record, are drawing attention half a century later as a new family of materials. Inorganic chemists at the time studied many layered compounds, but this was before research on two-dimensional materials such as graphene took off, so the potential of those structures was not fully explored. Some of these compounds, it turns out, can be classified as "MXenes," a group now considered important in materials science. Re-examining past research from a modern perspective has opened a path to discovering new materials.

On October 29, 2026, an international research team led by Linköping University in Sweden announced that it had identified previously overlooked MXene candidates by combining machine-learning analysis of databases with a review of past literature. The findings were published in the journal Science. The paper is titled "Data-driven discovery and rapid, direct synthesis of MXenes" (DOI: 10.1126/science.aeh5957). The team found 33 candidates that had been synthesized in the past but never recognized as MXenes, and used that knowledge to successfully synthesize 11 new layered compounds containing rare-earth elements. Including re-syntheses of known compounds, the number of multilayer MXenes obtained experimentally reaches 18.

This result does not mean that AI autonomously designed unknown materials and manufactured them instantly at room temperature. In practice, it was materials development that combined machine-learning structure classification, literature review by researchers, first-principles calculations, and self-propagating high-temperature synthesis (SHS). In the synthesis, a brief ignition triggers a high-temperature chemical reaction that proceeds without sustained external heating and finishes within minutes. The significance of the study lies in the full workflow: finding promising materials from past knowledge and then synthesizing them experimentally.

{
  "type": "bar",
  "title": "本研究における候補同定と合成実績の内訳",
  "unit": "種",
  "categories": ["多層MXene類似構造の同定数", "歴史的に未認識だった候補数", "実際に合成されたM2CT2相の総数", "うち新規希土類系の合成相数"],
  "series": [
    { "name": "物質数", "values": [56, 33, 18, 11] }
  ],
  "caption": "Science誌掲載論文（DOI: 10.1126/science.aeh5957）に基づくデータ分類",
  "source": "Linköping University / Science"
}

To understand these numbers, it is necessary to distinguish between candidates identified from databases and compounds actually synthesized in this study. The 33 overlooked candidates are substances that had already been synthesized; not all of them were newly made this time. The team re-evaluated past crystal structures and, building on that knowledge, synthesized a total of 18 multilayer phases. Of these, 11 are rare-earth compositions reported for the first time.

33 Unrecognized MXenes Found Among 56 Crystal Structures

MXene is a general term for two-dimensional materials made mainly of transition-metal carbides and nitrides, first reported in 2011 by researchers at Drexel University in the US. A typical MXene is several atomic layers thick, with fluorine, oxygen, hydroxyl groups, and other species bonded to its surface. Thanks to their excellent electrical conductivity and large specific surface area, MXenes have been studied for use in battery electrodes, electromagnetic shielding, catalysts, and more. Titanium-based MXenes in particular have been studied extensively, and the representative material Ti₃C₂Tₓ is known as a highly conductive 2D material.

Most conventional MXenes have been made from layered precursor crystals called MAX phases. MAX phases have a structure in which metal and other elements are stacked in a regular arrangement; selectively etching away certain atomic layers with hydrofluoric acid or similar agents leaves layers corresponding to MXene. Widening the gaps between layers and exfoliating them then yields thin, sheet-like material. However, this method takes time for precursor synthesis and etching, and the steps that handle strong acids require attention to safety and environmental impact.

This study focused on multilayer structures represented by the general formula \(\mathrm{M}_2\mathrm{XT}_2\). M is a metal element, X is mainly carbon or nitrogen, and T is a terminating element located on the surface of a layer made of the metal and carbon. In such structures, thin layers of M and X are stacked via the terminating elements, and if the layers can be separated, 2D sheets may be obtained. Importantly, some compounds with this structure were known long before the name MXene came into being.

The team used machine learning to survey crystal structure databases and cross-checked the results against the original research literature, arriving at 56 experimentally reported compounds with structures corresponding to multilayer MXenes. Of these, 33 are candidates that had not previously been recognized as a type of MXene. In older studies, such substances were treated as layered halide carbides and the like, and their properties had not been investigated sufficiently from the current 2D-materials perspective. By revisiting past crystallographic data, the team rediscovered a group of compounds that had been buried in existing materials research.

That said, machine learning did not design 33 new substances from scratch. The AI's main role was to extract, from a large number of crystal structures, candidates with atomic arrangements resembling MXenes. Confirming whether the candidates actually exist and what their crystal structures are required researchers to trace back to past papers and verify them. Repositioning known compounds within current materials classifications became the starting point for new synthesis research.

Machine Learning and Literature Review Widened the Search for Materials

The team's search method combined automated analysis of large databases with literature review by experts. Targeting inorganic crystal structures recorded in the Materials Project and the Open Quantum Materials Database (OQMD), they used machine-learning models to select candidates with atomic arrangements and layered structures characteristic of multilayer MXenes. By narrowing candidates using crystal symmetry, interatomic distances, and element placement, they found compounds that earlier research had not treated as MXenes.

On the other hand, the information registered in crystal structure databases is not necessarily complete. Crystal data imported from old literature can contain missing structural information or differences in notation, and the fact that a compound was experimentally synthesized may not be properly classified in the database. The team therefore checked the original academic papers and crystal structure measurements for the candidates extracted by machine learning. Compounds reported in Germany and elsewhere in the 1970s were also re-examined, and the research results of that era fed into this materials search.

The investigation revealed that the chemical composition of MXenes is more diverse than previously thought. Earlier research centered on transition metals such as titanium, vanadium, and molybdenum, but the candidates this time included many layered compounds containing rare-earth elements. In particular, structures incorporating scandium, yttrium, lanthanides, and similar elements at the M site may have electronic states and magnetic properties different from those of conventional representative MXenes.

The combinations of elements at the X and T sites also went beyond the range widely studied in conventional MXenes. As for terminating elements in particular, compounds incorporating not only fluorine but also chlorine and bromine were confirmed. Because the type of terminating element affects interlayer distance and electronic structure, it may be possible to design materials with different properties even from the same framework of metal and carbon. Based on these search results, the team chose halogen-terminated carbides containing rare-earth elements as new synthesis targets.

Self-Propagating High-Temperature Synthesis Directly Produced 18 Multilayer MXenes

Using the knowledge gained from database analysis, the team worked on direct synthesis of MXenes without relying on conventional etching. The method they adopted is called self-propagating high-temperature synthesis (SHS). In this approach, localized ignition energy is applied to a mixture of highly reactive raw materials, and the heat generated by the chemical reaction then sustains it. Even without prolonged external heating, an exothermic reaction propagating through the sample can form the target compound in a short time.

Specifically, metal powders, a carbon source, and a halogen-supplying raw material are combined, and the reaction is initiated by means such as brief resistive heating. After ignition, the exothermic reaction proceeds spontaneously and synthesis is completed without continuous external heating. According to the team's report, the reaction finishes within minutes. However, this does not mean the chemical reaction proceeds at room temperature; the reacting region reaches temperatures well above 1,000°C. The hallmark of the method is that a brief ignition triggers a high-temperature combustion reaction.

In its announcement on October 29, Linköping University also emphasized that materials can be synthesized in an extremely short time without sustained external heating. However, the short synthesis time the university describes does not mean the temperature during the reaction is low. Reducing the steps that maintain external heating must be distinguished from the reaction itself proceeding at low temperature. Nor can it be concluded that a short reaction time means low energy consumption across the entire manufacturing process.

The team first confirmed the method's effectiveness by directly synthesizing the known compounds Y₂CF₂ and Y₂CCl₂. They then synthesized, using the same method, bromine-terminated Y₂CBr₂, which had been reported in the past but not sufficiently recognized as an MXene. The ability to obtain layered compounds with different terminating elements—fluorine, chlorine, and bromine—directly suggests that composition can be controlled by changing the combination of raw materials.

The team then moved on to synthesis using praseodymium, neodymium, samarium, gadolinium, terbium, holmium, and thulium. As a result, they succeeded in synthesizing 11 rare-earth MXenes that had not been reported before. Including re-syntheses of known compounds, the multilayer MXenes synthesized in this study total 18. This adds a new group of materials using rare-earth elements to MXene research that had centered on transition metals.

The synthesized samples were examined in detail using X-ray diffraction (XRD), scanning transmission electron microscopy (STEM), and other techniques. The results confirmed a structure in which atomic layers of metal and carbon are stacked in a regular arrangement, consistent with structure predictions from first-principles calculations. However, the formation of the target crystal phase is not the same as the samples being free of impurities. For practical use, it will be necessary to raise the proportion of the target phase and maintain consistent quality regardless of synthesis conditions.

The new synthesis method may also reduce environmental impact. Because it does not require the conventional hydrofluoric acid etching step, it is expected to reduce the handling of strong acids and the use of rinse water. Linköping University also cites lower use of hazardous chemicals and water compared with conventional methods as an advantage. However, this study did not quantitatively evaluate the environmental impact of the entire manufacturing process. A life cycle assessment (LCA) covering raw material production, treatment of by-products and exhaust gases from the combustion reaction, and equipment operation is needed, and at this stage it cannot be said that an environmental advantage has been fully demonstrated.

Semiconducting and Magnetic Properties of Rare-Earth MXenes Predicted by Calculation

The newly synthesized rare-earth MXenes may have electronic and magnetic properties different from those of conventional materials. However, most of the properties presented in the paper are predictions from first-principles calculations, and not all have been directly confirmed by experiment. Confirming a material's crystal structure by actually synthesizing it is a separate step from showing that the material works as expected in an electronic device.

The team ran calculations using density functional theory (DFT) on 45 M₂CT₂ compositions containing yttrium and lanthanides. Evaluating thermodynamic stability, dynamic stability based on lattice vibrations, and mechanical stability while varying combinations of metal and terminating elements, they predicted that 43 compositions meet the stability criteria they set. However, stability in calculation does not guarantee that every material can be easily synthesized. In actual synthesis, the reactivity of raw materials, temperature conditions, competition with impurity phases, and other factors have an influence, so experimental verification is needed for each material.

Electronic structure calculations predicted that many rare-earth MXenes would be semiconductors with band gaps of roughly 1 to 2 eV. This contrasts with Ti₃C₂Tₓ, the representative titanium-based MXene, which generally exhibits metallic electrical conductivity. If they can be used as semiconductors, applications such as transistors and photodetectors are conceivable. In addition, certain cerium- and ytterbium-based compositions were predicted by calculation to show properties close to a "half-metal," being metallic for one electron spin while having an energy gap for the other.

For semiconducting properties, optical measurements on some of the synthesized samples were carried out in addition to theoretical calculations. For example, for Y₂CF₂, the optical band gap was estimated from ultraviolet-visible diffuse reflectance spectra, and a value roughly consistent with the calculated result was obtained. Such measurements support the idea that the electronic structure predicted by calculation is reflected in the actual material. However, measuring the optical band gap alone cannot confirm charge mobility or operating characteristics as a transistor.

Also noteworthy are the magnetic properties derived from the 4f electrons of rare-earth elements. Most lanthanides have localized 4f electrons, and interactions between their spins may produce magnetic order such as ferromagnetism or antiferromagnetism. The team's calculations likewise showed that the stable magnetic state differs depending on the element, with some compositions predicted to be ferromagnetic and others antiferromagnetic. Because differences in terminating elements change interlayer distance and electronic state, it may be possible to tune magnetic interactions even when using the same metal element.

Comparison with existing experimental results is also important for magnetism. For example, for Ho₂CF₂, which contains holmium, antiferromagnetic order with a Néel temperature of 3.7 K has been reported in past research, making it experimental knowledge relevant to this theoretical analysis. However, not all of the predicted magnetic orders for the new phases with various terminating elements have been confirmed by direct measurement. How far magnetic properties can be controlled by changing the terminating element will need to be verified through detailed magnetization measurements and magnetic structure analysis in the future.

These results indicate that rare-earth MXenes could be used as semiconducting and magnetic materials. Applications such as spintronics, which uses electron spin for information processing, and electronic devices with new magnetic functions are conceivable. However, what the paper demonstrated is mainly the synthesis of the materials, confirmation of their structures, and basic property measurements and theoretical predictions; practical devices such as magnetic memory or transistors were not prototyped. The next challenge is whether the properties expected from calculation can also be obtained in actual device structures and operating environments.

Single-Layer Exfoliation and Mass Production Remain Challenges for Practical Use

What matters about this result is that new material candidates were found in past research records and linked to actual synthesis. Combining machine-learning structure searches with literature review by researchers showed that the value of already known compounds can be reassessed from a different angle. Furthermore, the direct production of rare-earth multilayer MXenes by self-propagating high-temperature synthesis is an important step toward broadening the range of material compositions. However, several technical challenges remain before this basic research can be connected to industrial applications.

The first challenge is technology for separating the synthesized multilayer crystals into single or few-layer thin sheets. The materials obtained in this study are multilayer crystals in which atomic layers are stacked regularly, and the individual layers are not necessarily separated. To take advantage of MXene's large specific surface area in battery electrodes, water treatment membranes, sensors, and similar uses, the step of widening the interlayer gaps, extracting thin sheets, and dispersing them becomes important depending on the application. Whether the rare-earth MXenes obtained by this direct synthesis method can be stably exfoliated into single layers without damaging their structure has not been sufficiently demonstrated.

The second challenge is quality control in mass production. Heat transfer, reaction rate, and product uniformity differ greatly between igniting a small amount of raw material in a laboratory and continuously processing large quantities at industrial scale. While self-propagating high-temperature synthesis has the advantage of using the heat generated by the reaction itself, technology to stably control the high-temperature reaction and suppress impurity formation is essential. Judging actual manufacturing cost will also require evaluating raw material prices, yield of the target phase, equipment costs, and safety measures.

The third challenge is the environmental impact of manufacturing and the long-term stability of the materials. Being able to omit the etching step that uses strong acids is an advantage, but safety and environmental effects must be assessed, including the handling of halogen-containing raw materials and by-products used in combustion synthesis. The burden associated with mining and refining rare-earth metals, as well as the materials' oxidation and degradation characteristics, cannot be ignored either. In particular, structural changes on exposure to air and moisture are an important item to verify when considering applications in batteries and electronic devices.

In addition, it must be confirmed whether the semiconducting and magnetic properties predicted this time will translate into practical performance. For example, as battery materials, charge-discharge capacity and cycle life would be evaluated; for electronic devices, carrier mobility and operational stability; and for magnetic devices, the temperature dependence and controllability of magnetic order. This study did not demonstrate performance as application devices. Being able to make a material of a new composition should be clearly distinguished from that material outperforming existing ones.

This study showed that re-evaluating past crystal data with modern machine learning and materials science methods can open up a new research area. Identifying 33 overlooked candidates and synthesizing a total of 18 multilayer MXenes, including 11 new rare-earth ones, are its concrete results. Going forward, in addition to single-layer exfoliation and improved synthesis quality, measured property evaluation and the establishment of mass-production technology will be important. Whether the newly found group of materials develops into practical batteries and electronic devices will become clear through future research.